Human Activity Recognition using Deep Neural Network

Mishra, Piyush; Dey, Sourankana; Ghosh, Suvro Shankar; Seal, Dibyendu Bikash; Goswami, Saptarsi

doi:10.1109/icdse47409.2019.8971476

Cited by 9 publications

(4 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Analisis Komponen Utama (PCA) umum digunakan untuk menemukan jumlah data yang lebih kecil dan tidak berkorelasi untuk mengurangi dimensi fitur [1]. Teknik klasifikasi lain yang diurutkan berdasarkan perkembangan historis adalah K-Nearest Neighbor (KNN) classifier [1] dan SVM [1,2,7,10,13], Hyper-sphere Multi-class SVM [6], Naive-Bayes [9,10], Random Forest [10], Multilayer Perceptron (MLP) [7,10], Convolutional Neural Networks (CNN) [4,8,13], LSTM [12,14,15,16,17], ResNet [20], Transformer [21,22,23]. Kompleksitas komputasi dari algoritma-algoritma di atas dalam memprediksi satu sampel berdasarkan yang tertinggi hingga yang terendah adalah O(kN2d) untuk KNN Video input diekstraksi menjadi beberapa baris frame/gambar untuk analisis pada tahap preprocessing.…”

Section: Metode Penelitianunclassified

Sistem Peringatan Tersemat untuk Pengemudi Mengantuk

Sari¹,

Enriko²

2023

JRRE

View full text Add to dashboard Cite

Pendeteksian Driver Drowsiness (DDD) merupakan teknologi keselamatan kendaraan penting yang dirancang untuk mencegah kecelakaan akibat kantuk pengemudi. Dalam penelitian ini, pendekatan baru diajukan menggunakan model jaringan saraf konvolusi (CNN) ringan yang terdiri dari 44.853 parameter. Berkat ke ringanannya, model ini bekerja secara efisien bahkan pada perangkat dengan sumber daya terbatas. Hasil percobaan menunjukkan kinerja kompetitif model ini dibandingkan dengan model yang ada dengan ukuran input dan jumlah parameter yang lebih besar. Dalam hal akurasi, metode ini mencapai akurasi sebesar 92,06% pada dataset Curtin Emotion Wheels (CEW) yang mengesankan. Bahkan dalam kondisi pencahayaan yang sulit, performa model ini tetap luar biasa jika digabungkan dengan kamera termal. Secara khusus, model ini mencapai akurasi yang luar biasa sebesar 95,10% pada jarak kamera 0,3-meter dari wajah pengemudi. Selain itu, metode ini memiliki karakteristik kecepatan yang sangat baik, sehingga cocok digunakan pada perangkat tertanam. Kecepatan rata-rata perangkat Raspberry Pi 4 diperkirakan mencapai 5 frames per detik (FPS). Hal ini menunjukkan kepraktisan dan kelayakan penerapan pendekatan ini dalam skenario waktu nyata, yang semakin meningkatkan keselamatan pengemudi.

show abstract

Section: Metode Penelitianunclassified

Sistem Peringatan Tersemat untuk Pengemudi Mengantuk

Sari¹,

Enriko²

2023

JRRE

View full text Add to dashboard Cite

show abstract

“…In a study [4], traditional algorithms such as naïve Bayes, hidden Markov model (HMM), hidden semi Markov model (HSMM) and conditional random fields (CRFs) are compared with deep learning models on raw sensory data and validated that those deep learning models outperformed the best result by 40%. Similar deep learning approaches have been previously employed for recognition in [5], [6]. The concept of combining AdaBoost with other classifiers (C4.5, multilayer perceptron and logistic regression (LR)) was introduced in [7].…”

Section: Related Workmentioning

confidence: 99%

Health monitoring catalogue based on human activity classification using machine learning

Abdulhussein

Hassen

Gupta

et al. 2022

IJECE

View full text Add to dashboard Cite

In recent times, fitness trackers and smartphones equipped with different sensors like gyroscopes, accelerometers, global positioning system sensors and programs are used for recognizing human activities. In this paper, the results collected from these devices are used to design a system that can assist an application in monitoring a person’s health. The proposed system takes the raw sensor signals as input, preprocesses it and using machine learning techniques outputs the state of the user with minimum error. The objective of this paper is to compare the performance of different algorithms logistic regression (LR), support vector machine (SVM), k-nearest neighbor (k-NN) and random forest (RF). The algorithms are trained and tested with an original number of features as well as with transformed number of features (using linear discriminant analysis). The data with a smaller number of features is then used to visualize the high dimensional data. In this paper, each data point is mapped in the high dimensional data to two-dimensional data using t-distributed stochastic neighbor embedding technique. Overall, the first high dimensional data is visualized and compared with model’s performance with different algorithms and different number of coordinates.

show abstract

“…On the one hand, the recognition could be done by extracting handcrafted time-domain and frequency-domain features from raw signals, to feed classifiers like k-Nearest Neighbors (KNN), Random Forest (RF), and Deep Neural Network (DNN) [22,30,36]. Moreover, Random Subspace (RS) technique has been proposed to process Quaternions and Euler angles [37].…”

Section: Towards Human Activity Recognitionmentioning

confidence: 99%

“…The input of CNN and LSTM is raw data (time-series representing Euler angles). Besides, a large set of timedomain and frequency-domain features (proposed in [22,30,36]) were extracted from Euler angles to feed DNN (same architecture as in [22]), KNN ( =5), and RF (100 estimators). Since large feature vectors could mislead the machine, they were subject to a feature selection/dimensionality reduc-…”

Section: Comparison With State-of-the-art Techniquesmentioning

confidence: 99%